What is it that you see when you see? You see an object as a key, a man in a car as a passenger, some sheets of paper as a book. It is this word ‘as’ that must be mathematically formalised, on par with connectives “and”, “or”, “implies”, and “not”. …Until you do that, you will not get very far with your AI problem.
By Stanislaw Ulam, as cited in Bitwise: A Life in Code
Yesterday, I met two guys from Micronesia. They spoke in broken English, and we could communicate little directly. I asked him how far Micronesia is from the US; he said he’d been here in the US for three years. How do you say “hi” in Micronesian? Kaselehlie. I had to Google this word; I didn’t remember it at all. Why? The word is so unlike everything I’ve heard.
We build knowledge based on our existing knowledge. Every new piece of information needs to be connected to a part of currently existing knowledge. In some ways, this is how our neural networks work.
This picture from Obsidian, a note-taking app designed to arrange every new note related to a previous message based on a tag, conveys the idea practically.
Our current models either predict or classify. Either of these actions is not a complete representation of how human intelligence work. In fact, we are far away from understanding how human intelligence works!
Added on Sunday, July 17, 2022.